Effects of link perturbation on network modularity for community detections in complex network systems

Abstract

Community detection is of great significance in analyzing the network structures. However, real networks usually contain missing links and spurious interactions, which affect the accuracy of community detection results. In this paper, we aim to find out the regularity of the impact on community detection when links are deleted from or added to the network. To address this problem, we propose degree-related link perturbation (DRLP) methods for the tasks of both deleting and adding links, and the random perturbation methods are also be employed. Then, we evaluate the impact of perturbation methods on community detection and draw some conclusions. Finally, extensive experiments conducted on six real-world networks demonstrate the existence of the regularity. The perturbation of deleting and adding links can lead to continuous rise and decline of modularity, respectively, which is also instructive to change the results of community detection purposefully.